Research map / Image generation / Text-to-image diffusion
Rethink Sparse Signals for Pose-guided Text-to-image Generation
Wenjie Xuan, Jing Zhang, Juhua Liu, Bo Du, Dacheng Tao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add spatial-pose representation (ablation) | ControlNet · AP-10K | pose mAP: 44.25 -> 52.85 | Improved |
| add keypoint concept learning (ablation) | ControlNet · AP-10K | pose mAP: 44.25 -> 51.34 | Improved |
| combine both modules (ablation) | ControlNet · AP-10K | pose mAP: 44.25 -> 55.63 | Improved |
| add keypoint concept learning on top of spatial pose (ablation) | ControlNet · AP-10K | pose mAP: 52.85 -> 55.63 | Improved |
| add both modules (CLIP-Score cost) | ControlNet · AP-10K | CLIP-Score: 24.77 -> 23.86 | Hurt |
| use keypoint tokens at inference rather than removing them (ablation) | SP-Ctrl · AP-10K | pose mAP: 45.07 -> 51.34 | Improved |
| learnable spatial embedding module vs fixed RGB OpenPose embeddings (ablation) | ControlNet · AP-10K | pose mAP | Improved |
| random vs text-embedding initialization of E0 (ablation) | spatial-pose representation · AP-10K | pose mAP: 0.88% higher mAP | Improved |
| all-ones skeleton embedding vs interpolation (ablation) | spatial-pose representation · AP-10K | pose mAP: 0.86% higher | Improved |
| all-ones skeleton embedding vs learnable vectors (ablation) | spatial-pose representation · AP-10K | pose mAP: 1.96% higher | Improved |
| vary embedding dimension C' (ablation) | spatial-pose representation · AP-10K | pose mAP | No clear effect |
| apply heatmap loss at time steps 250-500 only (ablation) | keypoint concept learning · AP-10K | pose mAP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 5 of 8 shared results
- T2I-Adapter — T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024) · better on 6 of 6 shared results
- GRPose — GRPose: Learning Graph Relations for Human Image Generation with Pose Priors (AAAI 2025) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Follow Your Pose: Pose-Guided Text-to-Video Generation Using Pose-Free Videos (AAAI 2024 · similarity 0.86)
- Stable-Pose: Leveraging Transformers for Pose-Guided Text-to-Image Generation (NeurIPS 2024 · similarity 0.86)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.85)
- UniMC: Taming Diffusion Transformer for Unified Keypoint-Guided Multi-Class Image Generation (ICML 2025 · similarity 0.85)
- TrioPose: Native Triple-Stream Diffusion Transformers for Pose-Guided Text-to-Image Generation (NeurIPS 2026 · similarity 0.84)
- Motion Prompting: Controlling Video Generation with Motion Trajectories (CVPR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
